Start building agents — free.
A small, runnable taste of how AIS builds agents: four beginner patterns you can read in an afternoon and paste into your own project today. Pure Python 3.10+, zero dependencies, no API key — every file runs offline against a tiny mock model you swap for your real LLM client in one place.
Stage 0 of the agent-engineering ladder. No setup, no install, no required order.
Four patterns to start with
Every folder contains exactly two runnable files: the pattern (run it first — it’s the demo) and the test (run it second — it’s the proof). No setup, no install, no API key.
Prompt structure
Assemble a prompt from explicit parts — system, role, task, constraints, examples, output format — instead of one hopeful line.
f01-prompt-structure/
Structured output
Ask for JSON against a small schema, validate it with plain Python, and on invalid output send exactly one bounded repair naming the violation — then escalate.
f02-structured-output/
Tool calling
A tool registry with per-tool argument validation and an allowlist, where tools return errors as data and never raise into the loop.
f03-tool-calling/
Context basics
A context budget that assembles the model’s view each turn and trims oldest-first when it exceeds a token cap — never dropping the goal or open constraints.
f04-context-basics/
# no install, no dependencies, no API key cd f01-prompt-structure python prompt_structure.py # 1. watch the pattern work python test_prompt_structure.py # 2. prove it works on your machine # every folder is independent — no required order
Get the pack
Grab the pack right now with the direct download — no email required. Or drop your email to join the list for new releases and the occasional sharp note, and follow the daily Agent Foundations series for the reasoning behind each pattern.
Join the list for new patterns.
New releases, the occasional sharp note, and a copy of the pack by email. Free.
No spam. Unsubscribe anytime.
Ready for more?
Foundations is Stage 0 of the agent-engineering ladder. When these four patterns feel familiar, the paid libraries go deeper on what comes after.
Agent Foundations
Four beginner patterns — prompts, structured output, tool calling, context. You are here.
Loop Engineering
How one agent loop finishes, verifies, and stops. 13 tested patterns.
View the library →Graph Engineering
How many steps, tools, and agents coordinate. 10 tested patterns.
View the library →Agentic Systems Hardening
The defenses that keep it safe in production. 10 tested patterns.
View the library →